You are on page 1of 35

ECONOMIC GROWTH CENTER

YALE UNIVERSITY
P.O. Box 208269
New Haven, CT 06520-8269
http://www.econ.yale.edu/~egcenter/
CENTER DISCUSSION PAPER NO. 884
MICROECONOMIC FLEXIBILITY IN LATIN AMERICA
Ricardo J. Caballero
Massachusetts Institute of Technology
Eduardo M.R.A. Engel
Yale University
and
Alejandro Micco
Inter-American Development Bank
March 2004
Notes: Center Discussion Papers are preliminary materials circulated to stimulate discussions and
critical comments.
We thank Fernando Coloma and participants at the 7
th
Annual Conference of the Central Bank
of Chile for their comments.
This paper can be downloaded without charge from the Social Science Research Network
electronic library at: http://ssrn.com/abstract=512582
An index to papers in the Economic Growth Center Discussion Paper Series is located at:
http://www.econ.yale.edu/~egcenter/research.htm
Microeconomic Flexibility in Latin America
Ricardo J . Caballero, Eduardo M.R.A. Engel, and Alejandro Micco
Abstract
We characterize the degree of microeconomic inflexibility in several Latin American
economies and find that Brazil, Chile and Colombia are more flexible than Mexico and Venezuela.
The difference in flexibility among these economies is mainly explained by the behavior of large
establishments, which adjust more promptly in the more flexible economies, especially when
accumulated shocks are substantial. We also study the path of flexibility in Chile and show that it
declined in the aftermath of the Asian crisis. This decline can account for a substantial fraction of
the large decline in TFP-growth in Chile since 1997 (from 3.1 percent per year for the preceding
decade, to about 0.3 percent after that). Moreover, if it were to persist, it could permanently shave
off almost half of a percent from Chiles structural rate of growth.
Keywords: Microeconomic rigidities, creative-destruction, job flows, restructuring and
reallocation, productivity growth
JEL Codes: E2, J 2, J 6
1 Introduction
Although with varying degrees of success, Latin American economies have begun to leave behind
some of the most primitive sources of macroeconomic uctuations. Gradually, policy concern is
shifting toward increasing microeconomic exibility. This is a welcome trend since, by facilitating
the ongoing process of creative-destruction, microeconomic exibility is at the core of economic
growth in modern market economies.
But how poorly are these economies doing along this exibility dimension? Answering this
question requires measuring the important but elusive concept of microeconomic exibility. How
do we do this?
One way is to look directly at regulation, perhaps the main institutional factor hindering or
facilitating microeconomic exibility. In particular, there are extensive studies of labor market
regulation. Heckman and Pages (2000), for example, document that even after a decade of sub-
stantial deregulation [in most cases], Latin American countries remain at the top of the Job Security
list, with levels of regulation similar to or higher than those existing in the highly regulated South
of Europe. This is important work. However, in practice microeconomic exibility depends not
only on labor market regulation, but also on a wide variety of factors, including the technological
options and nature of the production process, the political environment, the efciency and biases
of labor courts, as well as cultural variables and accepted practices. Thus, while useful for eventual
policy formulation, studies of rules and regulation are unlikely to provide us with the big picture
of a countrys exibility any time soon understanding the complex interactions of different
regulations and environments is a valuable but very slow process.
At the other extreme, one can look at outcomes directly: How much factor reallocation do
we see in different countries and episodes? This is also a useful exercise. However, it is equally
incomplete since there is no reason to expect the same degree of aggregate ows in countries
facing different idiosyncratic and aggregate shocks. Hence it is always difcult to know whether
the observed reallocation is abnormally high or low, since the counterfactual is not part of the
statistic.
A third approach, which remedies some of the main weaknesses of the previous ones, is to
measure microeconomic exibility by the speed at which establishments reduce the gap between
their labor productivity and the marginal cost of such labor. Thus, we say an economy is inexible
at the microeconomic level if these gaps persist over time. Conversely, a very exible economy,
rm, or establishment, is one in which gaps disappear quickly due to prompt adjustment. This is
the approach we follow in this paper, extending a methodology developed in Caballero, Cowan,
Engel and Micco (2004) the main advantage of this methodology over conventional partial
1
adjustment estimates is its ability to use limited information efciently, correcting standard biases
often present when estimating such models. Our methodology also allows for nonlinearities and
state-dependent responses of employment to productivity gaps, as in Caballero and Engel (1993).
1
We use establishment level observations for all the Latin American economies for which we
had access to fairly reliable data: Chile, Mexico and, to a lesser extent, Brazil, Colombia and
Venezuela. All in all, about 140,000 observations.
In the rst part of the paper we document the main features of adjustment for these economies.
We nd that:
While more inexible than the US, on average (over time) Brazil, Colombia and Chile ex-
hibit a relatively high degree of microeconomic exibility with over 70 percent of labor
adjustment taking place within a year. Mexico ranks lower with about 60 percent of adjust-
ment within a year, and Venezuela is the most inexible of these economies, with slightly
over 50 percent of adjustment within a year.
With the only exception of Venezuela, in all our economies small establishments (below
the median number of employees) are substantially less exible than large establishments
(above the 75th percentile of employees). In Brazil, the former establishments close about
67 percent of their gap within a year, while the latter close about 81 percent. In Colombia,
68 and 79, respectively; in Chile 69 and 78; Mexico 56 and 61; and Venezuela 53 percent
for both.
It follows from the previous nding that it is primarily the behavior of large establishments
that is behind the substantial differences in exibility across some of the economies we study.
It may well be the case that large companies in Venezuela and Mexico are more insulated
from competitive pressures than their counterparts in Colombia, Chile and Brazil.
In all these economies there is evidence of an increasing hazard. That is, establishments
are substantially more exible with respect to large gaps than to small ones. This points
to the presence of signicant xed costs of adjustment, which may have a technological or
institutional origin.
The increasing hazard feature is particularly pronounced in large establishments in the rel-
atively more exible economies. In fact, most of the additional exibility experienced by
1
Note that our denition of microeconomic exibility refers to the speed at which establishments react to changing
conditions; not to whether the labor market is exible or not in responding to aggregate shocks. Thus, a labor market
regulation that makes the real wage rigid will result in a larger unemployment response to aggregate shocks that is,
it will exhibit macroeconomic inexibility yet this will not be part of our measure of microeconomic inexibility.
2
large establishments in these economies is due to their rapid adjustment when gaps get to be
large. For example, when gaps are below 25 percent in Chile, small establishments have an
adjustment coefcient of 0.50 while large ones have one of 0.51. For deviations above 25%,
on the other hand, small establishments have a coefcient of 0.79, while large establishments
have one of 0.93. The patterns are similar in Brazil and Colombia, yet less pronounced in
Mexico and Venezuela.
In the second part of the paper we specialize on Chile, which has the only long panel in our
sample, and explore the evolution of its microeconomic exibility over time. Our main ndings
are the following:
Microeconomic exibility in Chile experienced a signicant decline toward the end of our
sample (1997-99). From an average adjustment coefcient of 0.77 for the three years prior
to the Asian/Russian crisis episode, the coefcient fell to 0.69 in the aftermath of the crisis.
When the adjustment hazard is assumed to be constant, the decline in exibility appears
to be subsiding toward the end of the sample. However, this nding is lost and there is
no evidence of recovery once the hazard is allowed to be increasing. The reason for the
misleading conclusion with a constant hazard is that toward the end of the sample there is a
sharp rise in the share of establishments with large (negative) gaps, to which establishments
naturally react more under increasing hazards.
While it is too early to tell whether the decline we uncover is purely cyclical, or whether there
is something more structural going on, there are a few interesting observations to make:
a) Much of the decline in exibility is due to a decline in the exibility of large establish-
ments.
b) While the speed of response to negative gaps remained fairly constant, it is the speed
at which establishments adjust to shortages of labor that slowed down more dramati-
cally. This reluctance to hire may reect pessimism regarding future conditions not
captured in the contemporaneous gap. But this is unlikely to be the only factor since
otherwise we also should observe a rise in the speed of ring (for a given hazard).
2
c) Finally, the sharpest decline in exibility came from establishments in sectors that nor-
mally experience less restructuring, either because of smaller shocks or more techno-
logical and institutional inexibility. If either form of inexibility is responsible for
2
While we did see an increase in the speed of ring, as we argued above, this is accounted for by the interaction of
a prolonged contraction with an increasing hazard.
3
reduced restructuring, then the cost of the decline in exibility can be potentially very
large, as already inexible establishments spend signicant time away from their fric-
tionless optimum.
In the last part of the paper we explore a different metric for the degree of inexibility and its
economic impact. By impairing worker movements from less to more productive units, microe-
conomic inexibility reduces aggregate output and slows down economic growth. We develop a
simple framework to quantify this effect. Our ndings suggest that the aggregate consequences
of micro-inexibilities in Latin America are signicant. In particular, the impact of the decline in
microeconomic exibility in Chile following the Asian crisis is in itself large enough to account for
a substantial fraction of the decline in TFP-growth in Chile since 1997 (from an annual average of
3.1 percent for the preceding decade to about 0.3 percent after that). Moreover, if it were to persist,
it could permanently shave off almost half of a percent from Chiles structural rate of growth.
Section 2 presents the methodology while Section 3 describes the data. Section 4 characterizes
average microeconomic exibility in the Latin American economies in our data. Section 5 explores
the case of Chile in more detail, and describes the evolution of its index of exibility. Section 6
presents a simple model to map microeconomic inexibility into growth outcomes. Section 7
concludes and is followed by several appendices.
2 Methodology and Data
2.1 Overview
The starting point for our methodology is a simple adjustment hazard model, where the change in
the number of (lled) jobs in establishment i in sector j between time t 1 and t is a probabilistic
(at least to the econometrician) function of the gap between desired and actual (before adjustment)
employment:
e
i jt
=
i jt
(e

i jt
e
i jt1
), (1)
where e
i jt
and e

i jt
denote the logarithm of employment and desired employment, respectively.
The random variable
i jt
, which is assumed i.i.d. both across establishments and over time, takes
values in the interval [0, 1] and has mean and variance (1), with 0 1. The case =0
corresponds to the standard quadratic adjustment model, the case =1 to the Calvo (1983) model.
The parameter captures microeconomic exibility. As goes to one, all gaps are closed quickly
and microeconomic exibility is maximum. As decreases, microeconomic exibility declines.
4
Equation (1) also hints at two important components of our methodology: We need to nd
a measure of the employment gap, (e

i jt
e
i jt1
), and an estimation strategy for the mean of the
random variable
i jt
, . We describe both ingredients in detail in what follows. In a nutshell, we
construct estimates of e

i jt
, the only unobserved element of the gap, by solving the optimization
problem of the rm, as a function of observables such as labor productivity and a suitable proxy
for the average market wage. We estimate from (1), based upon the large cross-sectional size of
our sample and the well documented fact that there are signicant idiosyncratic components in the
realizations of the gaps and the
i jt
s.
An important aspect of our methodology is to nd an efcient method to remove xed effects
while, at the same time, avoiding the standard biases present in dynamic panel estimation.
3
The
model we develop also leads to a standard dynamic panel formulation, namely:
4
Gap
i jt
= (1)e

i jt
+(1)Gap
i jt1
+
i jt
. (2)
We report results for this specication as well, using dynamic panel techniques, in Table 12. They
are consistent with the estimates we obtain based on (1) and therefore provide a useful robustness
check. Yet they are considerably less precise. Thus our methodology may be viewed as an alter-
native, for the particular problem at hand, that uses data more efciently than standard dynamic
panel estimation techniques.
2.2 Details
Output and demand for establishment are given by:
y = a+e +h, (3)
p = d
1

y, (4)
where y, p, e, a, h, d denote rm output, price, employment, productivity, hours worked and
demand shocks, and is the price elasticity of demand. We let (1)/.
5
All variables are
in logs.
Firms are competitive in the labor market but pay wages that are increasing in the average
3
As documented, for example, in Arellano and Bond (1991).
4
The Gap below could be the gap before or after adjustments take place.
5
In order to have interior solutions, we assume > 1 and < 1.
5
number of hours worked, according to:
6
w = w
o
+(hh), (5)
where h is constant over time and interpreted below.
7
A key assumption is that rms only face adjustment costs when they change employment lev-
els, not when they change the number of hours worked.
8
It follows that the rms choice of hours
in every period can be expressed in terms of its current level of employment, by solving the corre-
sponding rst order condition (FOC) for hours.
In a frictionless labor market the rms employment level also satises a FOC for employ-
ment. Our functional forms then imply that the optimal choice of hours does not depend on the
employment level.
9
We denote the corresponding employment level by e and refer to it as the static
employment target.
10
The following relation between the employment gap and the hours gap then
follows:
e e =

1
(hh). (6)
This is the expression used by Caballero et Engel (1993). It is not useful in our case, since we
do not have information on worked hours. Yet the argument used to derive (6) also can be used to
express the employment gap in terms of the marginal labor productivity gap:
e e =

1
(v w
o
),
where v denotes marginal productivity, /() is decreasing in the elasticity of the marginal
wage schedule with respect to average hours worked, 1, and w
o
was dened in (5). This result
is intuitive: the employment response to a given deviation of wages from marginal product will
6
The expression below should be interpreted as a convenient approximation for:
w = k
o
+log(H

+),
with w
o
and determined by k
o
and .
7
To ensure interior solutions, we assume > and > .
8
For evidence on this see Sargent (1978) and Shapiro (1986).
9
A patient calculation shows that
h =
1

log
_

_
.
.
10
We have:
e =C+
1
1
[d +aw
o
],
with C a constant that depends on , , and .
6
be larger if the marginal cost of the alternative adjustment strategy changing hours is higher.
Also note that e e is the difference between the static target e and realized employment, not the
dynamic employment gap e

i jt
e
i jt
related to the term on the right hand side of (1). However, we
assume that demand, productivity and wage shocks follow a random walk.
11
We then have that e

i jt
is equal to e
i jt
plus a constant
t
.
12
It follows that
e

i jt
e
i jt1
=

1
j
_
v
i jt
w
o
i jt
_
+e
i jt
+
t
, (7)
where we have allowed for sector-specic differences in .
We estimate the marginal productivity of labor (v
i jt
) using output per worker multiplied by an
industry-level labor share, assumed constant over time.
Two natural candidates to proxy for w
o
i jt
are the average (across each industry, at a given point
in time) of either observed wages or observed marginal productivities. The former is consistent
with our assumption of a competitive labor market, the latter may be expected to be more robust in
settings with long-termcontracts and multiple forms of rewards, where the salary may not represent
the actual marginal cost of labor.
13
Our estimations were performed using both alternatives and we
found no discernible differences. This suggests that statistical power comes mainly from the cross-
section dimension, that is, from the well documented and large magnitude of idiosyncratic shocks
faced by establishments. In what follows we report the more robust alternative and approximate
w
o
by the average marginal productivity, which leads to:
e

i jt
e
i jt1
=

1
j
(v
i jt
v
jt
) +e
i jt
+
t
Gap
i jt
+
t
. (8)
The expression above ignores systematic variations in labor productivity that may occur across
establishments, which would tend to bias estimates of the speed of adjustment downward. In
Appendix A we provide evidence in favor of incorporating this possibility by subtracting from
(v
i jt
v
jt
) in (8) a moving average of relative productivity by establishment,

i jt
.
14
The resulting
11
Fron the preceding footnote if follows that it sufces that d +aw
o
follows a random walk.
12
In order to allow for variations in future expected growth rates of a and d, the constant is allowed to vary over
time.
13
While we have assumed a simple competitive market for the base salary (salary for normal hours) within each
rm, our procedure could easily accommodate other, more rent-sharing like, wage setting mechanisms (with a suitable
reinterpretation of some parameters, but not ).
14
Where

i jt

1
2
[(v
i jt1
v
jt1
) + (v
i jt2
v
jt2
)]. The alternative specication, with relative wages instead of
relative marginal productivities, leads to almost identical results.
7
expression for the estimated employment-gap is:
15
e

i jt
e
i jt1
=

1
j
(v
i jt

i jt
v
jt
) +e
i jt
+
t
Gap
i jt
+
t
, (9)
Finally, we estimate (related to the substitutability between hours worked and employment)
using
e
i jt
=

1
j
(v
i jt
v
jt
) +
t
+
it
+e

i jt
z
i jt
+
t
+
i jt
, (10)
where is a year dummy, e

i jt
is the change in the desired level of employment and z
it
(v
i jt

v
jt
)/1
j
). By assumption e

i jt
is i.i.d. and independent of lagged variables. In order to
avoid endogeneity and measurement error bias we estimate (10) using (w
i jt1
w
jt1
) as an
instrument for (v
i jt
v
jt
).
16
Table 1 reports the estimation results of (10) across the countries
in our sample.
17
We report estimates both with and without the one percent of extreme values for
the independent variable. For ease of comparison across countries, based on the estimates reported
in Table 1 we choose a common value of equal to 0.40.
2.3 Summary
Our methodology has three advantages when compared with previous specications used to es-
timate cross-country differences in speed of adjustment. First, it only requires data on nominal
output and employment level, two standard and well-measured variables in most industrial sur-
veys. Most previous studies on adjustment costs require measures of real output or an exogenous
measure of sector demand.
18
Second, it summarizes in a single variable all shocks faced by a rm.
This feature allows us to increase precision, and therefore the power of hypothesis testing, and to
study the determinants of the speed of adjustment using interaction terms. Finally, our approach
can be extended to incorporate non-linearities in the adjustment function. That is, the possibility
that the in (1) depend on the gap before adjustments take place. This feature also turns out to be
15
Where
j
is constructed using the sample median of the labor share for sector j across year and countries (Brazil,
Chile, Colombia, Mexico and Venezuela).
16
We lag the dependent variable because it is correlated with the error term, and we use lagged wages to instrument
lagged labor productivity to avoid measurement errors.
17
We do not have wage data for Brazil, so we cannot estimate the parameter for this country.
18
Abraham and Houseman (1994), Hammermesh (1993), and Nickel and Nunziata (2000)) evaluate the differential
response of employment to observed real output. A second option is to construct exogenous demand shocks. Although
this approach overcomes the real output concerns, it requires constructing an adequate sectorial demand shock for
every country. A case in point are the papers by Burgess and Knetter (1998) and Burgess et al (2000), which use
the real exchange rate as their demand shock. The estimated effects of the real exchange on employment are usually
marginally signicant, and often of the opposite sign than expected.
8
useful.
Summing up, in our basic setup we estimate the microeconomic exibility parameter from
e
i jt
= (Gap
i jt
+
t
) +
i jt
, (11)
where Gap
i jt
is proportional to the gap between marginal labor productivity and the market wage.
To correct for labor heterogeneity across establishments, a xed effect is also included in the gap-
measure. This xed effect is estimated by the average labor productivity in the two preceding
periods. As shown in Appendix A, the resulting estimator is unbiased (on average). It forces us to
discard only two time periods, and can adapt to slow time variations in heterogeneity.
3 Data and basic facts
This section describes the source and data used in the empirical analysis. These data are from
manufacturing censuses and surveys conducted by national statistical government agencies in ve
Latin American countries: Brazil, Chile, Colombia, Mexico and Venezuela. The variables used
in our analysis are nominal output, employment, total compensation and industry classication
within the manufacturing sector (ISIC at three digits). For the case of Chile, we also use capital
stock and a measure of cash ow dened as sales minus total input costs.
For Brazil, the data are from the Manufacturing Annual Survey (Pesquisa Industrial Anual)
conducted by the Instituto Brasileiro de Geograa e Estatstica. This survey started in 1967 but
experienced a severe methodological change in 1996, thus we only use observations from 1996
to 2000. In this, as well as in all other countries, we only include plants that existed during the
full period (continuous plants). In the case of Chile the data are from the Chilean Manufacturing
Census (Encuesta Nacional Industrial Anual) conducted by the Instituto Nacional de Estadsticas.
In principle, the surveys covers all manufacturing plants in Chile with more than ten employees
during the period 1979-97. In the empirical section we only use continuous plants during the
period 1985-97. We do not use the years before 1985 because they are characterized by large
macroeconomic shocks and structural adjustments that introduce too much noise and complications
to our methodology. For Colombia we use the Colombian Manufacturing Census (Encuesta Anual
Manufacturera y Registro Industrial) conducted by the Departamento Administrativo Nacional de
Estadsticas. The survey covers all manufacturing plants with more than twenty employees during
the period 1982-99. For plants with less than twenty employees only a random sample is covered.
Again, we only use continuous plants during the period 1992-99 due to a methodological change
9
in the survey in 1992.
For Mexico we use the Mexican Manufacturing Annual Survey (Encuesta Industrial Anual)
conducted by the Instituto Nacional de Estadstica, Geografa e Inform atica. The survey covers
a random sample of rms in the manufacturing sector during the period 1993-2000. Finally, for
Venezuela the data are from the Manufacturing Survey (Encuesta Industria Manufacturera) con-
ducted by the Instituto Nacional de Estadistica. The survey covers all plants with more than 50
employees and it has a yearly random sample for plants with less than 50 employees. Due to
changes in the methodology, we only are able to follow rms during the 1995-1999 period.
Table 2 presents the number of observations per size bracket (measured by the number of
employees) for each of the ve countries, for the sample period at hand. The coverage of plants by
size differs across countries. Chile and Colombia have the largest coverage of small plants (less
than 50 employees), whereas Venezuelas survey mainly covers large establishments.
In table 3 we compute the average job creation and job destruction for each country. In addi-
tion we report the simple average over time of net change in employment and the excess turnover
(i.e., the sum of job ows net of the change in employment due to cyclical factors). All statistics
are dened following Davis et al. (1996). It is already apparent in these numbers that microeco-
nomic exibility in these countries is limited: they are of the same order of magnitude of those
of developed economies which presumably need less restructuring than catching-up emerging
economies and substantially below economies such as Taiwan.
19
4 Microeconomic Flexibility
In this section we report our average (over time) exibility ndings. The basic results are reported
in Table 4. All of our regressions include year-dummies, d
t
. That is, for each country, we estimate
(we drop the sector j subscript):
e
it
= d
t
+Gap
it
+
it
. (12)
The rst apparent result is that microeconomic exibility is more limited in our economies than
in the very exible US. In the latter, estimates of using annual data are much closer to one.
20
Although comparisons must be interpreted with caution since the samples differ in number
of observations, time-periods, establishments demographics, etc., there is a discernible pattern.
19
See e.g., Caballero and Hammour (2000) and references therein.
20
For example, Caballero, Engel and Haltiwanger (1997) nd a quarterly for US manufacturing exceeding 0.4,
which implies an annual of approximately 0.90.
10
Within the region, Brazil, Colombia and Chile exhibit a relatively high degree of microeconomic
exibility with over 70 percent of labor adjustment taking place within a year. Mexico ranks lower
with about 60 percent of adjustment within a year, and Venezuela is the most inexible of these
economies, with slightly more than 50 percent of adjustment within a year.
Lending support to our earlier motivation for adopting our approach in constructing a broad
measure of microeconomic inexibility, our ranking is essentially uncorrelated with the ranking
obtained by Heckman and Pages (2000) and Botero et al. (2003) based on measuring labor market
regulations (see Table 5). For example, and in contrasts to our results, the Botero et al. (2003)
index of job security places Venezuela at a level of exibility similar to that of Brazil and Chile,
and Colombia as signicantly more exible than all of the above.
21
Table 6 reports the results from repeating estimation of regression (12), but conditioning on
whether establishments are small or large. The former are dened as those with a number of
employees below the median in the preceding year, large ones are those above the 75th percentile
in number of employees (also in the preceding year).
In all our economies but Venezuela, small rms are substantially less exible than large estab-
lishments. In Brazil, the former close about 67 percent of their gap within a year, while the latter
close about 81 percent. In Colombia, 68 and 79, respectively; in Chile 69 and 78; Mexico 56 and
61; and Venezuela 53 percent for both.
It also follows from this table that it is primarily the behavior of large establishments that
explains the substantial differences in exibility across some of these economies. Again, this need
not come from differences in labor market regulation and hence it would not be captured by
such indices but it could also reect, for example, barriers to entry or social objectives assigned
to large rms.
In addition to splitting by size, Table 7 splits observations by the magnitude of the employment-
gap. Small gaps are dened as gaps of less than 25 percent, in absolute value, while large ones are
for gaps above 25 percent. That is, we re-estimate (12) for each country-size/size-of-gap combina-
tion ( jsg):
e
i jsgt
= d
jsgt
+
jsg
Gap
i jsgt
+
i jsgt
. (13)
There are several signicant conclusions that follow from this table:
1. In all the economies we study there is evidence of an increasing hazard.
22
That is, establish-
ments are substantially more exible with respect to large gaps than to small ones. This hints
21
Also, according to the Heckman and Pages (2000) index, the most exible countries in our sample are Brazil and
Mexico; not Chile and Colombia as suggested by our index.
22
See Caballero and Engel (1993) for a description of increasing hazard models and their aggregate implications.
11
at the presence of signicant xed costs (increasing returns) in the adjustment technology.
These xed costs may have a technological origin, as when there are strong complementar-
ities in production or xed proportion with sunk capital, or institutional, as when dismissals
require approval by a government agency or are likely to be litigated in court.
2. The increasing hazard feature is particularly pronounced in large establishments in the rela-
tively more exible economies. This does not mean that these rms face larger xed costs
than the same establishments in less exible economies. Quite the opposite, since they still
adjust more frequently than their counterparts in inexible economies. It means that the
benets of adjustments overcome xed costs sooner in large establishments in exible econ-
omies and that there are more elements of randomness (i.e., not correlated with the size of
the gap) in the adjustment decisions of large establishments in inexible economies.
3. In fact, most of the additional exibility experienced by large establishments in the more
exible Latin American economies is due to their rapid adjustment when gaps get to be very
large (over 25 percent). For example, both small and large establishments have an adjustment
coefcient of approximately 0.50 for gaps below 25% in Chile. For large deviations, on the
other hand, small establishments have a coefcient of 0.79, while large establishments have
one of 0.93. The patterns are similar in Brazil and Colombia, and less pronounced in Mexico
and Venezuela.
In conclusion, there is evidence of microeconomic inexibility in the Latin American econ-
omies, and in some cases, such as Mexico and Venezuela, the problem is quite severe. Studies
based only on quantifying job ows would be unable to detect either of these facts: Gross job
ows are comparable in magnitude to those in the US, and across all the economies we study,
or yield the wrong ranking (e.g., Chile would be the second most inexible of these economies,
according to the excess reallocation numbers presented in Table 3); the same remark applies to
studies solely based on studying labor markets regulation.
23
We also nd that allowing for an increasing hazard is important: There is clear evidence of
increasing hazards, especially for large establishments in the more exible economies. To a sub-
stantial extent, more inexible economies seem to be those where large imbalances go uncorrected
for sustained periods of time. Conversely, large establishments in the more exible economies
seldom tolerate (or can afford to tolerate) large microeconomic imbalances.
23
Of course there is plenty of merit and usefulness in such studies. Our remarks only refer to our attempt of
measuring a broad concept of microeconomic exibility.
12
5 The Evolution of Flexibility
Has microeconomic exibility improved over time? Unfortunately, we only count with a long
time dimension for the case of Chile. In what follows we specialize our analysis to this case, and
conclude that the answer to this question is negative. Quite the opposite, exibility has declined
signicantly since the Asian crisis.
All our results in this section are obtained from running variants of the regression:
e
i jt
= [
0jt
+
1 j
{|Gap
i jt
| > 0.25}+
2 j
{Gap
i jt
< 0.05}]Gap
i jt
+
+d
1j
{|Gap
i jt
| >0.25} + d
2 j
{Gap
i jt
<0.05} +
i jsgt
, (14)
where we include, but do not report, constants, time and group (e.g., |Gap
i jt
| > 0.25) dummies.
The results of these variants are reported in Table 8.
Figure 1 plots the path of the
0jt
s, with their mean subtracted. The solid lines represent
the results for all rms, the dashed lines those for large rms, and the dotted lines those for small
rms. Ahigh value represents an upward shift in the adjustment hazard. We focus on the shift in the
hazard itself as an index of exibility rather than on the average speed of adjustment, because in the
realistic increasing hazard context the latter depends on the endogenous path of the cross section.
When the hazard is constant, its shift also represents an equal shift in the speed of adjustment.
When the hazard is increasing, on the other hand, the mapping from a vertical shift in the hazard
to a change in the average speed of adjustment is not one-for-one, since the interactions with the
cross sectional distribution of gaps complicates the mapping.
Column 1 in Table 8 and the continuous line in the upper panel of Figure 1 show the results
for the constant hazard case. Under this assumption, the index of exibility exhibited uctuations
in the second half of the 1980s and early 1990s, eventually settled at a fairly high value in the mid
90s, but then declined sharply during the 1997-99 period. From an average adjustment coefcient
of 0.77 for the three years prior to the Asian/Russian crisis episode, this coefcient fell to 0.69 in
the aftermath of the crisis.
Note also that in this case the decline in exibility appears to be subsiding toward the end of
the sample. However columns 4 and 7 in Table 8, and the continuous lines in the middle and
lower panels of Figure 1, show that this nding is lost and there is no evidence of recovery once
the hazard is allowed to be nonlinear. The reason for the misleading conclusion with a constant
hazard is that toward the end of the sample there is a sharp rise in the share of establishments with
large negative gaps (see Figure 2), to which establishments naturally react more under increasing
13
hazards.
24
That is, the average speed of adjustment rises even if the hazard does not change, due
to substantial negative gaps accumulated by a large number of establishments.
While it is too early to tell whether this decline in microeconomic exibility we uncover is
purely cyclical, or whether there is something more structural going on, there are a few interesting
observations we can make at this time. We begin by noting that the remaining columns in Table 8
and series in Figure 1 show that much of the decline in exibility is due to a decline in the exibility
of large establishments (as measured by their lagged employment).
Continuing with the characterization of the decline in microeconomic exibility, Table 9 shows
that while the speed of response to negative gaps remained fairly constant, it is the speed at which
establishments adjust to shortages of labor that slowed down more dramatically.
25
This reluctance
to hire may reect pessimism respect to future conditions not captured in the current gap. But
this is unlikely to be the only factor since otherwise we also should observe a rise in the speed
of ring, which we do not. In fact, the increasing hazard nature of the adjustment hazard partly
explains the asymmetry seen in the decline of the speed of adjustment with respect to positive and
negative gaps. As we mentioned above, since there was a substantial number of establishments that
developed large negative gaps (excess labor) during the slowdown, the increasing hazard implied
that their adjustment did not slow down as much as the decline in the average speed of adjustment.
However, Table 10 illustrate that the sharpest decline in exibility came from establishments
in sectors that normally experience less restructuring, either because of smaller shocks or more
technological and institutional inexibility. Normal restructuring for high and low restructuring
sectors is measured by the excess reallocation above/below median in Chile prior to 1997.
26
If it
is not shocks but inexibility that explains the ranking, then the cost of the increase in exibility
can be potentially very large, as already inexible establishments spend signicant time away from
their frictionless optimum.
In conclusion, while we cannot pinpoint to a specic reason for why microeconomic exibility
declined toward the end of the 1990s, we clearly identied such a decline. Moreover, we found
that the increasing nature of the hazard is important to show that the recovery in average exibility
toward 1999 does not seem to correspond to a real increase in exibility. Instead, it simply reects
the interaction between an increasing hazard and a depressed phase of the business cycle. Flexi-
bility declined in 1997 and remained down until the end of our sample, particularly so for large
establishments. We also found that the decline in exibility is more pronounced in sectors that
24
Where large negative gaps are gaps smaller than 0.25 and large positive gaps are gaps larger than 0.25.
25
Between 1994-96 and 1997-99, the latter fell from 0.86 to 0.71, while the former fell from 0.75 to 0.71.
26
Similar results are obtained when sectors are classied according to the excess reallocation in the corresponding
US sectors (a sort of instrumental variables for technological factors).
14
normally restructure less. If the latter is a consequence of larger adjustment costs (technological or
institutional), then their relative slowdown is worrisome since the cost of reducing their restructur-
ing further is particularly large. In the next section we turn to gauging some of the potential costs
of microeconomic inexibility.
6 Gauging the Costs of Microeconomic Inexibility
By impairing worker movements from less to more productive units, microeconomic inexibility
reduces aggregate output and slows down economic growth. In this section we develop a simple
framework to quantify this effect. Any such exercise requires strong assumptions and our approach
is no exception. Nonetheless, our ndings suggest that the costs of microeconomic inexibilities in
Latin America are signicant. In particular, the impact of the decline in microeconomic exibility
in Chile following the Asian crisis accounts for a substantial fraction of the large decline in TFP-
growth in Chile since 1997 (from an annual average of 3.1 percent for the preceding decade to
about 0.3 percent after that). Moreover, if it were to persist, it could permanently shave off about
0.4 percent from Chiles structural rate of growth.
6.1 Model
Consider a continuum of establishments, indexed by i, that adjust labor in response to productivity
shocks, while their share of the economys capital remains xed over time. Their production
functions exhibit constant returns to (aggregate) capital, K
t
, and decreasing returns to labor:
Y
it
= B
it
K
t
L

it
, (15)
where B
it
denotes plant-level productivity and 0 < < 1. The B
it
s follow geometric random
walks, that can be decomposed into the product of a common and an idiosyncratic component:
logB
it
b
it
= v
t
+v
I
it
,
where the v
t
are i.i.d. N (
A
,
2
A
) and the v
it
s are i.i.d. (across productive units, over time and with
respect to the aggregate shocks) N (0,
2
I
). We set
A
=0, since we are interested in the interaction
between rigidities and idiosyncratic shocks, not in Jensen-inequality-type effects associated with
aggregate shocks.
The price-elasticity of demand is > 0. Aggregate labor is assumed constant and set equal to
15
one. We dene aggregate productivity, A
t
, as:
A
t
=

B
it
L

it
di, (16)
so that aggregate output, Y
t


Y
it
di, satises
Y
t
= A
t
K
t
.
Units adjust with probability in every period, independent of their history and of what other
units do that period.
27
The parameter that captures microeconomic exibility is . Higher values
of are associated with a faster reallocation of workers in response to productivity shocks.
Standard calculations show that the growth rate of output, g
Y
, satises:
28
g
Y
= sA, (17)
where s denotes the savings rate (assumed exogenous) and the depreciation rate for capital.
Consider now what happens when microeconomic exibility decreases from
0
to
1
. Aggre-
gate productivity decreases, reecting slower reallocation of workers from less to more productive
units. Indeed, from (16) we have that :
A =

B
it
L

it
di,
where L

it
denotes the difference between the value of L

it
for the new value of and the value it
would have had under the old . A tedious, but straightforward calculation relegated to Appendix
B shows that:
A
_
1

1
_
A
0
,
with
=
(2)
2(1)
2
(
2
I
+
2
A
),
and = (1)/.
Using (17) to get rid of A
0
yields our main result:
g
Y
(g
Y,0
+)
_
1

1
_
, (18)
27
More precisely, whether unit i adjusts at time t is determined by a Bernoulli random variable
it
with probability
of success , where the
it
s are independent across units and over time.
28
Here we use that g
A
= 0, since we assumed
A
= 0.
16
where g
Y,0
denotes the growth rate of output before the change in .
We choose parameters to apply (18) as follows: The mark-up is set at 20%. Parameters g
Y,0
,
I
and
A
are set at their average values for Chile over the 198796 period, namely 7.9%, 19% and
4%. We also set = 6%. The microeconomic exibility parameters are set at their average values
during 1994-96 and 1997-99 for large establishments,
29
since they concentrate most production.
Fromthis exercise we conclude that the reduction in exibility has reduced structural output growth
by 0.4%. This permanent cost is due to the effect of reduced productivity on capital accumulation.
One must add to this the initial direct effect of a decline in productivity on output growth,
30
which
amounts to 2.7 percent. The sum of these two structural costs is very relevant. As mentioned
earlier, it can account for a signicant share of the decline in Chilean TFP growth from an annual
average of 3.1 percent during the decade preceding the Asian crisis to 0.3 during the 1997-99
period.
Going back to the average results presented in Section 3, Table 11 reports the potential gain
in structural growth that each country could obtain from raising microeconomic exibility to US
levels. Our estimates indicate that, on the low end, Chile and Colombia would have an initial
gain in the range between 2 and 4% and a permanent increase in the structural rate of growth of
approximately 0.3%. On the high end, Venezuela would see an initial gain of 22.2%, even the
impact on its growth rate is less pronounced, due to it having had the lowest growth rate in our
sample. By contrast, Mexico could expect an initial gain of 7.4% and an impressive permanent
rise of growth of 0.7%, while the corresponding percentages for Brazil are 5.0 and 0.43. These
numbers are large. We are fully aware of the many caveats that such ceteris-paribus comparison
can raise, but the point of the table is to provide an alternative metric of the potential signicance
of observed levels of inexibility in our region.
7 Concluding Remarks
There is the nagging feeling among policymakers and observers that the microeconomic structure
of the Latin American economies is rather inexible, and that this is a signicant obstacle to
growth. Not surprisingly, pro-exibility structural reforms are high in most of the countries in the
region.
29
Equal to 0.688 and 0.892, respectively, see Table 8.
30
This is equal to:
A
A
0

_
1

1
_
.
17
Despite this widespread belief, there is very little in terms of formal and systematic evidence,
both on the extent of inexibility and on its costs. The data and methodological obstacles to
produce this evidence are signicant.
In this paper we collect extensive data sets for several Latin American countries. We then
develop a methodology suitable to extract an answer to the inexibility questions from these data
sets.
Our estimates conrm the above fears. Microeconomic inexibility is signicant and very
costly in our region. Moreover, in Chile, where we could measure the time path of exibility with
some precision, the trend does not seem to be pointing in the right direction. Our initial estimates
suggest that the decline in exibility observed at the end of the 1990s, if it were to persist, could
shave off near half of a percent from Chiles potential growth rate.
18
References
[1] Abraham, K. and Houseman, S. (1994). Does Employment Protection Inhibit Labor Market
Flexibility: Lessons From Germany, France and Belgium. In: R.M. Blank, editor. Protection
Versus Economic Flexibility: Is There A Tradeoff? Chicago, United States:University of
Chicago Press.
[2] Arellano, M. and S.R. Bond (1991). Some Specication Tests for Panel Data: Montecarlo
Evidence and an Application to Employment Equations, Review of Economic Studies, 58,
277-298.
[3] Botero, J., S. Djankov, R. La Porta, F. Lopez-de-Silanes and A. Shleifer (2003).The Regu-
lation of Labor. Harvard mimeo
[4] Burgess, S. and M. Knetter (1998). An International Comparison of Employment Adjust-
ment to Exchange Rate Fluctuations. Review of International Economics, 6(1): 151-163.
[5] Burgess, S., M. Knetter and C. Michelacci (2000). Employment and Output Adjustment in
the OECD: A Disaggregated Analysis of the Role of Job Security Provisions, Economica
67, 419-435.
[6] Caballero, R., K. Cowan, E. Engel and A. Micco (2003). Microeconomic Inexibility and
Labor Regulation: International Evidence, Mimeo, October 2003.
[7] Caballero R. and E. Engel (1993). Microeconomic Adjustment Hazards and Aggregate Dy-
namics. Quarterly Journal of Economics. 108(2): 359-83.
[8] Caballero, R., E. Engel and J. Haltiwanger (1997). Aggregate Employment Dynamics:
Building from Microeconomic Evidence, American Economic Review, 87 (1), 115137,
March 1997.
[9] Caballero, R. and M. Hammour (2000). Creative Destruction and Development: Institu-
tions, Crises, and Restructuring, Annual World Bank Conference on Development Economics
2000, 213-241.
[10] Calvo, G (1983). Staggered Prices in a Utility Maximizing Framework. Journal of Mone-
tary Economics, 12, 383-98.
[11] Davis, S., J. Haltiwanger and S. Schuh (1996), Job Creation and Destruction, Cambridge,
Mass.: MIT Press.
19
[12] Heckman, J and C. Pages (2000). The Cost of Job Security Regulation: Evidence from Latin
American Labor Markets. Inter-American Development Bank Working Paper 430.
[13] Hamermesh, D.(1993). Labor Demand, Princeton University Press.
[14] Nickell, S., and L. Nunziata (2000). Employment Patterns in OECD Countries, Center for
Economic Performance Dscussion Paper 448.
[15] Sargent, T. (1978), Estimation of Dynamic Labor Demand under Rational Expectations, J.
of Political Economy, 86, 100944.
[16] Shapiro, M.D., The Dynamic Demand for Capital and Labor, Quarterly Journal of Eco-
nomics, 101, 51342.
20
APPENDIX
A Estimating
Our starting point is (1) in the main text, where for simplicity we ignore sectors and time-variation
in the targets drift:
e
i,t
=
i,t
(e

it
e
i,t1
), (19)
with
i,t
: i.i.d., with mean and variance (1 ); [0, 1]. We denote by z
it
the gap after
period t adjustments; that is, z
i,t
e

it
e
it
. We assume
e

i,t
= e

A,t
+
i,t
,
with e

A,t
i.i.d. with mean
A
and variance
2
A
and
i,t
i.i.d. independent from the e

i,t
s, with
zero mean and variance
2
I
.
Given an integer M = 2, 3, ... we dene:
z
M
i,t
=
1
M
M

k=1
z
i,tk
. (20)
The central idea is that with plant-specic xed effects (e.g., systematic differences in labor force
composition) we do not observe the zs implicit on the r.h.s. of (19), but only observe the difference
z
i,t
z
M
i,t
(since the xed effects cancel out once we subtract z
M
). We therefore x t and estimate (19)
with z z
M
on the r.h.s. instead of z. One advantage of this approach is that the estimated values of

t
do not vary with the length of the time period considered, as is the case when estimating xed
effect using the time-average over the whole sample.
Denote
2
t
Var[z
i,t
], where the variance is calculated over i, keeping t xed. Also denote by

t
the OLS estimator of
t
, again keeping t xed and regressing over i. A calculation from rst
principles then shows that for M = 2 we have:
E[

t
] =
t
_
1+

2
t1

2
t2
4Var(z
i,t
z
M
i,t
+l
i,t
)
_
, (21)
with

2
t
=
1
t

t
[+(1)
t
]
_
[1(1)
t
]Var(e
i,t
) +
(2
t
1)

t
(e
A,t
)
2
_
, (22)
where e
A,t
denotes the average (over i) of e
i,t
.
It follows from (21) that, the time average of the estimates for
t
will be unbiased, since on
average
2
t1
is equal to
2
t2
. Of course, for any particular t, the estimator may be biased. Yet
the expression in (22) can be used to correct the bias in (21), since it expresses the bias in terms
21
of observables. We calculated the actual bias for the Chilean data and it is rather small, for all
periods.
Expressions analogous to (21) can be obtained for values of M larger than 2 and, surprisingly,
the average unbiasedness result described above holds only for M =2.
31
An additional advantage
of the M = 2 case is that, if the xed effect changes slowly over time, then the added precision
associated with larger values of M comes at the expense of a larger bias due to time-varying xed-
effects. In this sense, M = 2 provides a good compromise.
B Gauging the Costs
Here we show that, for the model in Section 6:
A
A
0

_
1

1
_
, (23)
with
=
(2)
2(1)
2
(
2
I
+
2
A
), (24)
and = (1)/.
The intuition is easier if we consider the following, equivalent, problem. The economy consists
of a very large and xed number of rms (no entry or exit). Production by rm i during period t
is Y
i,t
= A
i,t
L

i,t
,
32
while (inverse) demand for good i in period t is P
i,t
=Y
1/
i,t
, where A
i,t
denotes
productivity shocks, assumed to follow a geometric random walk, so that
logA
i,t
a
i,t
= v
A
t
+v
I
i,t
,
with v
A
t
i.i.d. N (0,
2
A
) and v
I
i,t
i.i.d. N (0,
2
I
). Hence a
i,t
follows a N (0,
2
T
), with
2
T
=
2
A
+
2
I
.
We assume the wage remains constant throughout.
In what follows lower case letters denote the logarithm of upper case variables. Similarly,

-variables denote the frictionless counterpart of the non-starred variable.


Solving the rms maximization problem in the absence of adjustment costs leads to:
l

i,t
=

1
a
i,t
, (25)
and hence
y

i,t
=
1
1
a
i,t
. (26)
Denote by Y

t
aggregate production in period t if there were no frictions. It then follows from (26)
31
Of course, as M tends to innity the estimator is (asymptotically) unbiased, without the need of averaging over
time.
32
That is, we ignore hours in the production function.
22
that:
Y

i,t
= e
a
i,t
Y

i,t1
, (27)
with 1/(1 ), Taking expectations (over i for a particular realization of v
A
t
) on both sides
of (27) and noting that both terms being multiplied on the r.h.s. are, by assumption, independent
(random walk), yields
Y

t
= e
v
A
t
+
1
2

2
I
Y

t1
, (28)
Averaging over all possible realizations of v
A
t
(these uctuations are not the ones we are interested
in for the calculation at hand) leads to
Y

t
= e
1
2

2
T
Y

t1
,
and therefore for k = 1, 2, 3, ...:
Y

t
= e
1
2
k
2

2
T
Y

tk
. (29)
Denote:
Y
t,tk
: aggregate Y that would attain in period t if rms had the frictionless optimal levels
of labor corresponding to period t k. This is the average Y for units that last adjusted k
periods ago.
Y
i,t,tk
: the corresponding level of production of rm i in t.
From the expressions derived above to follows that:
Y
i,t,t1
Y

i,t
=
_
L

i,t1
L

i,t
_

= e
a
i,t
,
and therefore
Y
i,t,t1
= e
a
i,t
Y

i,t1
.
Taking expectations (with respect to idiosyncratic and aggregate shocks) on both sides of the latter
expression (here we use that a
i,t
is independent of Y

i,t1
) yields
Y
t,t1
= e
1
2

2
T
Y

t1
,
which combined with (29) leads to:
Y
t,t1
= e
1
2
(1
2
)
2
T
Y

t
.
A derivation similar to the one above, leads to:
Y
i,t,tk
= e
a
i,t
+a
i,t1
+...+a
i,tk+1
Y

tk
,
23
which combined with (29) gives:
Y
t,tk
= e
k
Y

t
, (30)
with dened in (24).
Assuming Calvo-type adjustment with probability , we decompose aggregate production into
the sum of the contributions of cohorts:
Y
t
= Y

t
+(1)Y
t,t1
+(1)
2
Y
t,t2
+. . .
Substituting (30) in the expression above yields:
Y
t
=

1(1)e

t
. (31)
It follows that the production gap, dened as:
Prod. Gap
Y

t
Y
t
Y

t
,
is equal to:
Prod. Gap =
(1)(1e

)
1(1)e

. (32)
A rst-order Taylor expansion then shows that, when || << 1:
Prod. Gap
(1)

. (33)
Subtracting this gap evaluated at
0
from its value evaluated at
1
, and noting that this gap differ-
ence corresponds to A/A
0
in the main text, yields (23) and therefore concludes the proof.
24
Table 1: ESTIMATING
COUNTRY: Colombia Chile Mexico Venezuela

with extreme values: 0.414 0.460 0.372 0.336


(0.035) (0.028) (0.033) (0.108)

without extreme values: 0.394 0.495 0.365 0.317


(0.035) (0.037) (0.037) (0.118)
Observations: 20,268/20,065 21,149/20,938 27,752/27,475 2,906/2,877
Robust standard errors in parenthesis.
Table 2: DESCRIPTIVE STATISTICS I
COUNTRY: Brazil Colombia Chile Mexico Venezuela
Observations: 42,525 27,440 24,450 37,384 4,950
Establishments: 8,505 3,430 1,630 4,673 990
Employment (% obs.):
(0 , 50): 15.9 45.1 56.7 21.0 9.9
[50 , 100): 28.5 22.8 17.9 21.4 31.5
[100 , 250): 28.9 19.5 15.4 29.4 33.7
250: 26.6 12.7 9.9 28.2 24.9
Period: 1996-2000 1992-1999 1985-1999 1993-2000 1995-1999
Employment reports the percentage of observations with employment below 50, between 50 and 100,
between 100 and 250, and larger than 250. Only continuous plants are considered.
Table 3: DESCRIPTIVE STATISTICS II
COUNTRY: Brazil Colombia Chile Mexico Venezuela
Employment: 2,555,035 461,441 169,813 1,214,776 233,746
Net Change: 0.024 0.013 0.021 0.018 0.023
Job Creation: 0.074 0.072 0.080 0.071 0.069
Job Destruction: 0.098 0.086 0.059 0.053 0.091
Reallocation: 0.173 0.158 0.139 0.123 0.160
Excess Reallocation: 0.135 0.124 0.099 0.086 0.125
Period: 1997-2000 1993-1999 1986-1999 1994-2000 1996-1999
Quantities reported are yearly averages over the sample period. Detion of all variables follows Davis et
al. (1996).
25
Table 4: AVERAGE FLEXIBILITY ESTIMATES
COUNTRY: Brazil Colombia Chile Mexico Venezuela
Gap: 0.701 0.722 0.724 0.581 0.539
(0.004) (0.005) (0.005) (0.004) (0.014)
R-squared: 0.50 0.53 0.50 0.47 0.37
Observations: 25,260 20,375 20,979 27,757 2,941
Period: 1998-2000 1995-1999 1988-1999 1995-2000 1997-1999
Robust standard errors in parenthesis. All estimates in this table are signicant at the 1% level. All regres-
sions have year dummies. All estimates based on one regression per country, using all available observations.
Observations corresponding to extreme values (0.5% in right tail and 0.5% in left tail) of regressors excluded.
Table 5: COMPARING FLEXIBILITY MEASURES
COUNTRY: Brazil Colombia Chile Mexico Venezuela
Job Security Index (Heckman and Pages, 2000): 3.04 3.79 3.38 3.16 4.54
Job Security Index (Botero et al., 2003): 0.69 0.31 0.62 0.71 0.64
Excess Reallocation: 0.135 0.124 0.099 0.086 0.125
Microeconomic exibility index (this paper): 0.701 0.722 0.724 0.581 0.539
Flexibility is decreasing in the index for the rst two measures, and increasing for the remaining two measures.
Since yearly values for 19901999 are available for the Heckman-Pages index (this is not the case for the remaining
indices), the numbers reported for this index are the average over the sample period (years before 1990 are proxied by
the 1990 value, and years after 1999 by the 1999 value).
26
Table 6: AVERAGE FLEXIBILITY ESTIMATES BY PLANT SIZE
COUNTRY
Plant Size Brazil Colombia Chile Mexico Venezuela
Gap: Small 0.670 0.675 0.685 0.561 0.529
(0.006) (0.007) (0.007) (0.006) (0.020)
Large 0.808 0.790 0.783 0.607 0.529
(0.009) (0.010) (0.010) (0.007) (0.026)
R
2
: Small 0.47 0.52 0.49 0.44 0.35
Large 0.57 0.56 0.54 0.53 0.39
Obs.: Small 12,560 10,087 10,404 13,784 1,469
Large 6,340 5,131 5,265 7,008 741
Period: 1998-2000 1995-99 1988-99 1995-2000 1997-99
Small: below 50th percentile of the lagged employment distribution. Large: above the 75th percentile
of the lagged employment distribution. Robust standard errors in parenthesis. All estimates in this table
are signicant at the 1% level. All regressions have year dummies. Observations corresponding to extreme
values (0.5% in right tail and 0.5% in left tail) of regressor excluded.
27
Table 7: AVERAGE FLEXIBILITY ESTIMATES BY PLANT SIZE AND GAP SIZE
COUNTRY
Brazil Colombia Chile Mexico Venezuela
Plant Size Gap Size
Gap: Small Small 0.473 0.440 0.499 0.330 0.275
(0.010) (0.010) (0.009) (0.009) (0.033)
Large 0.722 0.752 0.790 0.626 0.570
(0.013) (0.012) (0.016) (0.010) (0.031)
Large Small 0.541 0.551 0.513 0.418 0.222
(0.011) (0.014) (0.013) (0.010) (0.044)
Large 0.870 0.890 0.927 0.682 0.540
(0.018) (0.020) (0.023) (0.015) (0.040)
R
2
: Small Small 0.21 0.22 0.27 0.14 0.08
Large 0.56 0.65 0.65 0.57 0.41
Large Small 0.28 0.29 0.29 0.26 0.06
Large 0.64 0.65 0.68 0.68 0.40
Obs.: Small Small 9,204 7,493 8,844 9,812 886
Large 3,356 2,594 1,560 3,972 583
Large Small 4,903 4,052 4,342 5,729 441
Large 1,437 1,079 923 1,279 300
Period 1998-2000 1995-99 1988-99 1995-2000 1997-99
Plant size can be small (below 50th percentile of the lagged employment distribution) or large (above the 75th
percentile of the lagged employment distribution). Gap size can be small (absolute value less than 0.25) or large
(absolute value larger than 0.26). Robust standard errors in parenthesis. All estimates in this table are signicant at the
1% level. All regressions have year dummies. Observations corresponding to extreme values (0.5% in right tail and
0.5% in left tail) of regressors excluded.
28
Table 8: EVOLUTION OF FLEXIBILITY: CHILE 198799
1 2 3 4 5 6 7 8 9
Constant hazard Increasing (and asymmetric) hazard
Plant size: all small large all small large all small large
Gap 87: 0.745 0.742 0.782 0.490 0.514 0.537 0.343 0.384 0.365
(0.030) (0.036) (0.068) (0.030) (0.038) (0.064) (0.030) (0.039) (0.063)
Gap 88: 0.674 0.707 0.716 0.424 0.481 0.445 0.272 0.344 0.270
(0.031) (0.041) (0.059) (0.031) (0.040) (0.058) (0.031) (0.040) (0.060)
Gap 89: 0.776 0.714 0.854 0.533 0.504 0.564 0.381 0.377 0.381
(0.038) (0.042) (0.054) (0.034) (0.043) (0.054) (0.035) (0.043) (0.055)
Gap 90: 0.677 0.656 0.765 0.441 0.478 0.488 0.274 0.326 0.289
(0.031) (0.039) (0.072) (0.030) (0.039) (0.068) (0.032) (0.041) (0.072)
Gap 91: 0.731 0.688 0.806 0.501 0.503 0.578 0.335 0.362 0.374
(0.033) (0.053) (0.058) (0.032) (0.050) (0.055) (0.034) (0.051) (0.058)
Gap 92: 0.740 0.705 0.758 0.520 0.522 0.503 0.359 0.380 0.302
(0.039) (0.063) (0.065) (0.036) (0.058) (0.063) (0.038) (0.062) (0.064)
Gap 93: 0.706 0.640 0.812 0.492 0.474 0.547 0.322 0.327 0.347
(0.034) (0.047) (0.066) (0.032) (0.046) (0.060) (0.033) (0.047) (0.065)
Gap 94: 0.730 0.656 0.913 0.515 0.487 0.639 0.345 0.339 0.443
(0.036) (0.050) (0.071) (0.035) (0.049) (0.066) (0.036) (0.050) (0.070)
Gap 95: 0.775 0.743 0.907 0.547 0.569 0.641 0.370 0.415 0.434
(0.034) (0.048) (0.072) (0.032) (0.044) (0.065) (0.033) (0.046) (0.069)
Gap 96: 0.808 0.706 0.856 0.577 0.531 0.582 0.402 0.378 0.386
(0.035) (0.055) (0.059) (0.034) (0.054) (0.056) (0.035) (0.055) (0.059)
Gap 97: 0.686 0.648 0.667 0.469 0.495 0.395 0.301 0.346 0.206
(0.033) (0.043) (0.073) (0.032) (0.042) (0.072) (0.034) (0.046) (0.074)
Gap 98: 0.669 0.614 0.667 0.425 0.446 0.377 0.242 0.285 0.168
(0.040) (0.051) (0.095) (0.038) (0.051) (0.091) (0.040) (0.052) (0.092)
Gap 99: 0.705 0.655 0.712 0.418 0.455 0.367 0.250 0.309 0.172
(0.034) (0.045) (0.076) (0.035) (0.048) (0.075) (0.038) (0.050) (0.080)
Gap(|Gap| > .25): 0.371 0.295 0.407 0.479 0.410 0.508
(0.016) (0.023) (0.031) (0.016) (0.023) (0.032)
Gap(Gap < .05): 0.095 0.172 0.012
(0.031) (0.420) (0.062)
|Gap| > .25: 0.002 0.027 0.023 0.004 0.019 0.012
(0.004) (0.006) (0.009) (0.005) (0.007) (0.010)
Gap < .05: 0.093 0.097 0.087
(0.003) (0.004) (0.007)
R
2
: 0.50 0.49 0.54 0.53 0.51 0.57 0.55 0.54 0.59
Plant size can be small (below 50th percentile of the lagged employment distribution) or large (above the 75th percentile of
the lagged employment distribution). Robust standard errors in parenthesis. All regressions have year dummies. Observations
corresponding to extreme values (0.5% in right tail and 0.5% in left tail) of regressors excluded.
29
Table 9: EVOLUTION OF FLEXIBILITY AND ASYMMETRIC HAZARDS
Gap (Gap < .05)
Year Coeff. St. Error Coeff. St. Error No. Obs.
1987 0.689 0.030 0.227 0.062 1300
1988 0.720 0.030 0.079 0.058 1216
1989 0.729 0.033 0.155 0.061 1248
1990 0.702 0.036 0.016 0.060 1155
1991 0.815 0.036 0.097 0.061 1153
1992 0.752 0.035 0.061 0.067 1151
1993 0.721 0.037 0.034 0.064 1124
1994 0.831 0.039 0.135 0.066 1073
1995 0.891 0.036 0.152 0.060 1134
1996 0.859 0.039 0.040 0.063 1139
1997 0.710 0.039 0.028 0.062 1146
1998 0.734 0.046 0.078 0.069 1144
1999 0.698 0.052 0.031 0.070 1252
Simple Average: 0.758 0.002
Table 10: EVOLUTION OF FLEXIBILITY AND EX-ANTE RESTRUCTURING
High Restructuring Low Restructuring
Year Coeff. St. Error No. Obs. Coeff. St. Error No. Obs.
1987: 0.745 0.024 902 0.749 0.030 709
1988: 0.750 0.023 898 0.552 0.029 712
1989: 0.824 0.023 904 0.698 0.031 705
1990: 0.704 0.025 911 0.640 0.026 706
1991: 0.722 0.023 902 0.748 0.030 710
1992: 0.722 0.025 908 0.768 0.031 709
1993: 0.786 0.024 909 0.575 0.027 713
1994: 0.767 0.025 913 0.689 0.029 711
1995: 0.765 0.023 904 0.788 0.030 717
1996: 0.824 0.024 906 0.788 0.029 705
1997: 0.722 0.026 912 0.634 0.027 702
1998: 0.723 0.026 911 0.580 0.029 705
1999: 0.733 0.027 895 0.664 0.029 700
Simple Average: 0.753 0.682
30
Table 11: GAINS FROM ACQUIRING US-TYPE FLEXIBILITY
COUNTRY: Brazil Colombia Chile Mexico Venezuela

I
(%): 27. 6 25.8 19.3 24.1 38.1
g
Y,0
(%): 2.7 2.7 6.6 3.5 2.0
Additional Growth Upon Impact (%): 5.0 3.8 2.1 7.4 22.2
Increase in Growth Rate (%): 0.43 0.33 0.27 0.70 0.18
Table 12: FLEXIBILITY ESTIMATES BASED ON (2)
COUNTRY: Brazil Chile Mexico Venezuela
Gap: 0.855 0.675 0.592 0.401
(0.048) (0.034) (0.037) (0.184)
Observations: 8,322 17,631 18,368 968
Period: 1998-2000 1988-1999 1995-2000 1997-1999
Robust standard errors in parenthesis. The dependent variable is the change in the gap (after adjustments).
Second and third lag are used as instruments. All estimates in this table are signicant at the 1% level, with
the exception of Venezuela, which is signicant at the 5% level. All estimates based on one regression per
country, using all available observations. Colombia was not included because we did not have access to the
data. All regressions that consider more than one year (Chile and Mexico) use year dummies. Observations
corresponding to extreme values (0.5% in right tail and 0.5% in left tail) of regressors excluded.
31
Figure 1:
1986 1988 1990 1992 1994 1996 1998 2000
0.2
0.1
0
0.1
0.2
1986 1988 1990 1992 1994 1996 1998 2000
0.2
0.1
0
0.1
0.2
1986 1988 1990 1992 1994 1996 1998 2000
0.2
0.1
0
0.1
0.2
Constant hazard
Increasing hazard
Increasing and asymmetric hazard
32
Figure 2:
1986 1988 1990 1992 1994 1996 1998 2000
0
0.05
0.1
0.15
0.2
0.25
Fraction of exteme negative gaps
1986 1988 1990 1992 1994 1996 1998 2000
0
0.05
0.1
0.15
0.2
0.25
Fraction of extreme positive gaps
33

You might also like